[
https://issues.apache.org/jira/browse/DRILL-2993?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14537406#comment-14537406
]
Daniel Barclay (Drill) commented on DRILL-2993:
-----------------------------------------------
> I don't think disabling throttling is the right solution. The semantics of
> the RPC layer expect all messages to be consumed and ack'd. Is there some
> reason this isn't possible?
By "disabling throttling" I mean stopping throttling (that is, telling the
server to stop holding off on sending queued messages and to just go ahead and
just send them all), so that the server sends all the queued messages normally
(presumably addressing your concern about ACKs) and that the JDBC
implementation layer now gets all those messages (so its existing
post-cancelation consumption code can consume them (including freeing
allocations), presumably addressing your concern about consuming messages).
Given that clarification, does the approach seem correct, or is it still
problematic?
> SQLLine hangs when we cancel a query in the middle of displaying results
> ------------------------------------------------------------------------
>
> Key: DRILL-2993
> URL: https://issues.apache.org/jira/browse/DRILL-2993
> Project: Apache Drill
> Issue Type: Bug
> Components: Execution - Flow
> Reporter: Rahul Challapalli
> Assignee: Daniel Barclay (Drill)
> Attachments: DRILL-2993.1.patch.txt
>
>
> git.commit.id.abbrev=8c706e6
> The data set contains 1 million records. I cancelled the below query after
> displaying around ~400,000 records. All subsequent queries fail
> {code}
> 0: jdbc:drill:schema=dfs_eea>select * from `mobile.json`;
> .....
> .....
> Cancel after displaying 400,000 records
> 0: jdbc:drill:schema=dfs_eea>use dfs.drillTestDir;
> ....
> This hangs indefinitely
> {code}
> Since the data is large, I did not attach it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
